""" NamedValueAccess provides functions, a mix-in class and a wrapper class all for accessing Python objects by named attributes. You can use which ever of the three approaches best suites your needs and style. NOTES If Python provided a root class 'Object' in the same tradition as other OOP languages such as Smalltalk, Objective-C and Java, then we could dispense with the global functions and simply stick with the mix-in. TO DO * The mix-in's valueForKey() could be out of slight alignment with the function, since they have different implementations. However, the test cases pass for both right now. * Should the valueForKey() function provide for caching of bindings in the same manner than the mix-in does? If not, should the mix-in allow an option to *not* cache bindings? * hasValueForKey() function? (We already have a method in the mix-in) * valuesForNames() in the mix-in: * Change parameter 'keys' to 'names' * Use NoDefault instead of None in the parameters * Revisit doc string and test cases * Docs: More improvs to doc strings. * Testing: increase coverage * Rename? class NamedValueAccess+ible: * Benchmarking: Set this up in a new file: Testing/BenchNamedValueAccess.py so we can experment with caching vs. not and other techniques. PAST DESIGN DECISIONS * Only if a name binds to a method is it invoked. Another approach is to invoke any value that is __call__able, but that is unPythonic: If obj.foo is a class or a function then obj.foo gives that class or function, not the result of invoking it. Method is the only convenience we provide, because that's one of the major points of providing this. CREDIT Chuck Esterbrook <echuck@mindspring.com> Tavis Rudd <tavis@calrudd.com> """ import types import string, sys from time import time from MiscUtils import NoDefault # if technique is zero, use bound methods in the _kvGetBindings cache, otherwise use unbound # @@ 2000-05-31 ce: after additional testing we can probably scorge the technique=0 allowance technique = 1 ## Exceptions ## class NamedValueAccessError(LookupError): pass class ValueForKeyError(NamedValueAccessError): pass class NamedValueAccess: """ This class is intended to be ancestor class such that you can say: from NamedValueAccess import * age = someObj.valueForName("age") name = someObj.valueForName("info.fields.name") This can be useful in setups where you wish to textually refer to the objects in a program, such as an HTML template processed in the context of an object-oriented framework. Keys can be matched to either methods or ivars and with or without underscores. valueForName() can also traverse bona fide dictionaries (DictType). You can safely import * from this module. Only the NamedValueAccess class is exported (other than typical things like string and sys). There is no __init__() method and never will be. You can run the test suite by running this module as a program. You'll see the terms 'key' and 'name' in the class and its documentation. A 'key' is a single identifier such as 'foo'. A name could be key, or a qualified key, such as 'foo.bar.boo'. Names are generally more convenient and powerful, while key-oriented methods are more efficient and provide the atomic functionality that name-oriented methods are built upon. From a usage point of view, you normally just use the 'name' methods and forget about the 'key'. @@ 2000-05-21 ce: This class causes problems when used in WebKit for logging. Perhaps circular references? Involving self? Having to do with methods bound to their objects? @@ 2000-03-03 ce: document ivars @@ 2000-04-24 ce: Some classes like UserDict need to use getitem() instead of getattr() and don't need to deal with _bindingForGetKey(). @@ 2000-05-31 ce: Rename this class to NamedValues, NamedValueAccess, ValuesByName @@ This class probably needs to be in MiscUtils, as it's being used in that way while MiddleKit was intended for "enterprise/business objects". """ # # Accessing values by key # def hasValueForKey(self, key): """ Returns true if the key is available, although that does not guarantee that there will not be errors caused by retrieving the key. """ return self._bindingForGetKey(key)!=None def valueForKey(self, key, default=NoDefault): """ Suppose key is 'foo'. This method returns the value with the following precedence: 1. Methods before non-methods 2. Public attributes before private attributes More specifically, this method then returns one of the following: * self.foo() * self._foo() * self.foo * self._foo ...or default, if it was specified, otherwise invokes and returns result of valueForUnknownKey(). Note that valueForUnknownKey(), normally returns an exception. See valueForName() which is a more advanced version of this method that allows multiple, qualified keys. """ binding = self._bindingForGetKey(key) if not binding: if default is NoDefault: return self.valueForUnknownKey(key, default) else: return default if type(binding) is types.MethodType: # @@ 2000-05-07 ce: come to a decision on exception handling for key errors # try: if technique: result = binding(self) else: result = binding() # except: # @@ 2000-02-18: Improve next line with exception info # raise NamedValueAccessError, 'Caught exception while accessing key (%s). Exception is %s' % (key, sys.exc_info()) return result else: return getattr(self, binding) def hasValueForName(self, keysString): try: value = self.valueForName(keysString) except NamedValueAccessError: return 0 return 1 def valueForName(self, keysString, default=None): """ Returns the value for the given keysString. This is the more advanced version of valueForKey(), which can only handle single names. This method can handle 'foo', 'foo1.foo2', 'a.b.c.d', etc. It will traverse dictionaries if needed. """ keys = string.split(keysString, '.') return self.valueForKeySequence(keys, default) def valueForKeySequence(self, listOfKeys, default=None): # @@ 2000-02-18: document return _valueForKeySequence(self, listOfKeys, default) def valuesForNames(self, keys, default=None, defaults=None, forgive=0, includeNames=0): """ Returns a list of values that match the given keys, each of which is passed through valueForName() and so could be of the form 'a.b.c'. keys is a sequence. default is any kind of object. defaults is a sequence. forgive and includeNames is a flag. If default is not None, then it is substituted when a key is not found. Otherwise, if defaults is not None, then it's corresponding/parallel value for the current key is substituted when a key is not found. Otherwise, if forgive=1, then unknown keys simply don't produce any values. Otherwise, if default and defaults are None, and forgive=0, then the unknown keys will probably raise an exception through self.valueForUnknownKey() although that method can always return a final, default value. if keys is None, then None is returned. If keys is an empty list, then None is returned. Often these last four arguments are specified by key. Examples: names = ['origin.x', 'origin.y', 'size.width', 'size.height'] obj.valuesForNames(names) obj.valuesForNames(names, default=0.0) obj.valuesForNames(names, defaults=[0.0, 0.0, 100.0, 100.0]) obj.valuesForNames(names, forgive=0) @@ 2000-03-04 ce: includeNames is only supported when forgive=1. It should be supported for the other cases. It should be documented. It should be included in the test cases. """ if keys is None: return None if len(keys) is 0: return [] results = [] if default is not None: results = map(lambda key, myself=self, mydefault=default: myself.valueForName(key, mydefault), keys) elif defaults is not None: if len(keys) is not len(defaults): raise NamedValueAccessError, 'Keys and defaults have mismatching lengths (%d and %d).' % (len(keys), len(defaults)) results = map(lambda key, default, myself=self: myself.valueForName(key, default), keys, defaults) elif forgive: results = [] uniqueObject = 'uni' + 'que' for key in keys: value = self.valueForName(key, uniqueObject) if value is not uniqueObject: if includeNames: results.append((key, value)) else: results.append(value) else: # no defaults, no forgiveness results = map(lambda key, myself=self: myself.valueForName(key), keys) return results def setValueForKey(self, key, value): # @@ 2000-02-18: naming might be weired here with args reversed """ Suppose key is 'foo'. This method sets the value with the following precedence: 1. Public attributes before private attributes 2. Methods before non-methods More specifically, this method then uses one of the following: @@ 2000-03-04 ce: fill in ...or invokes handleUnknownSetKey(). """ raise NotImplementedError # @@ 2000-03-04 ce def resetKeyBindings(self): # @@ 2000-02-18 document this method if hasattr(self, '_kvGetBindings'): self._kvGetBindings = {} # # Errors # def valueForUnknownKey(self, key, default): raise NamedValueAccessError, key #def handleUnknownSetKey(self, key): # raise NamedValueAccessError, key # # Private # def _bindingForGetKey(self, key): """ Bindings are cached. Bindings are methods or strings. """ # Make _kvGetBindings dictionary if we don't have one if not hasattr(self, '_kvGetBindings'): self._kvGetBindings = {} # Return the binding if we already have one if self._kvGetBindings.has_key(key): return self._kvGetBindings[key] # No binding, so we have to look for the key found = None # set to what we find # Try plain old key if hasattr(self, key): found = getattr(self, key) #print '0: found = ', found, type(found) if type(found) is not types.MethodType: found = key elif technique: found = getattr(self.__class__, key) self._kvGetBindings[key] = found #print '1: found = ', found, type(found) # Try _key only if we didn't find a method called key if type(found) is not types.MethodType: underKey = '_' + key if hasattr(self, underKey): underAttr = getattr(self, underKey) if found==None: if type(underAttr) is types.MethodType: if technique: value = getattr(self.__class__, underKey) else: value = underAttr else: value = underKey found = self._kvGetBindings[key] = value else: if type(underAttr) is types.MethodType: if technique: underAttr = getattr(self.__class__, underKey) found = self._kvGetBindings[key] = underAttr #print '2: found = ', found, type(found) return found class NamedValueAccessWrapper(NamedValueAccess): """ This provides a wrapper around an existing object which will respond to the methods of NamedValueAccess. By using the wrapper, you can stick with objects and methods such as obj.valueForName('x.y') (as opposed to functions like valueForName()) and refrain from modifying the existing class hierarchy with NamedValueAccess. Example: wrapper = NamedValueAccessWrapper(obj) print wrapper.valueForName('manager.name') """ def __init__(self, object): self._object = object def hasValueForKey(self, key): try: value = self.valueForKey(ley) except NamedValueAccessError: return 0 else: return 1 def valueForKey(self, key, default=NoDefault): return valueForKey(self._object) def valueForName(self, key, default=NoDefault): return valueForName(self._object) # # Private # def _valueForKeySequence(obj, listOfKeys, default=None): """ This is a recursive function used to implement NamedValueAccess.valueForKeySequence. Besides supporting inheritors of NamedValueAccess, this function also supports dictionaries, which is why it's not found in the class. """ # @@ 2000-02-18: Optimize by specifying index instead of making new list if type(obj) is types.DictType: try: value = obj[listOfKeys[0]] except: # @@ 2000-03-03 ce: this exception should be more specific. probably nameerror or indexerror if default is None: raise NamedValueAccessError, 'Unknown key (%s) in dictionary.' % listOfKeys[0] else: return default else: value = obj.valueForKey(listOfKeys[0], default) if len(listOfKeys)>1: return _valueForKeySequence(value, listOfKeys[1:], default) else: return value def _dict_valueForKey(obj, key, default=NoDefault): """ Returns the value for a given key of the dictionary-like object. This is a private, custom function built in support of valueForKey(). """ try: value = obj[key] except AttributeError, e: # We attempt to pass only on exceptions caused # by obj not responding to __getitem__. Any # other exceptions generated get raised up. substring = "instance has no attribute '__getitem__'" if e.args[0][-len(substring):]==substring: if default is NoDefault: return None else: return else: raise except KeyError, e: if e.args[0]==key: if default is NoDefault: raise ValueForKeyError, key else: return default else: # If we get here, then the KeyError is deeper in the # implementation of obj[key] raise else: return value def valueForKey(obj, key, default=NoDefault): """ Returns the value of the object named by the given key. Suppose key is 'foo'. This method returns the value with the following precedence: 1. Methods before non-methods 2. Attributes before keys (__getitem__) 3. Public things before private things (private being denoted by a preceding underscore) More specifically, this method returns one of the following: * obj.valueForKey(key) # only if the method exists * obj.foo() * obj._foo() * obj.foo * obj._foo * obj['foo'] * obj.valueForUnknownKey(key) * default # only if specified If all of these fail, a ValueForKeyError is raised. NOTES * If the object provides a valueForKey() method, that method will be invoked to do the work. * valueForKey() works on dictionaries and dictionary-like objects. * valueForUnknownKey() provides a hook by which objects can delegate or chain their keyed value access to other objects. The key and default arguments are passed to it and it should generally respect the typical treatment of the the default argument as found throughout Webware and described in the Style Guidelines. * See valueForName() which is a more advanced version of this function that allows multiple, qualified keys. """ # We only accept strings for keys assert type(key) is types.StringType # Use obj.valueForKey() if it is available valueForKeyMeth = getattr(obj, 'valueForKey', None) if valueForKeyMeth: return valueForKeyMeth(key, default) attr = None method = None value = None unknown = 0 if type(obj) is types.DictType: if default is NoDefault: try: return obj[key] except KeyError: raise ValueForKeyError, key else: return obj.get(key, default) else: try: klass = obj.__class__ except AttributeError: raise AttributeError, '__class__ obj type=%r, obj=%r' % (type(obj), obj) method = getattr(klass, key, None) if not method: underKey = '_' + key method = getattr(klass, underKey, None) if not method: attr = getattr(obj, key, NoDefault) if attr is NoDefault: attr = getattr(obj, underKey, NoDefault) if attr is NoDefault: getitem = getattr(obj.__class__, '__getitem__', None) if getitem: try: value = getitem(obj, key) except KeyError: unknown = 1 # if value is not NoDefault: # return value if not unknown: if method: return method(obj) if attr is not NoDefault: return attr # Use obj.valueForUnknownKey() if it is available valueForUnknownKey = getattr(obj, 'valueForUnknownKey', None) if valueForUnknownKey: return valueForUnknownKey(key, default) if default!=NoDefault: return default else: raise ValueForKeyError, key def valueForName(obj, name, default=NoDefault): """ Returns the value of the object that is named. The name can use dotted notation to traverse through a network/graph of objects. Since this function relies on valueForKey() for each individual component of the name, you should be familiar with the semantics of that notation. Example: valueForName(obj, 'department.manager.salary') """ names = string.split(name, '.') for name in names: obj = valueForKey(obj, name, default) if obj is default: return obj # 2001-04-19 ce: I suppose the above technique could result in # the default being returned prematurely if it was part of the # chain of names. Well, that's just the way it goes for now. return obj # Beef up UserDict with the NamedValueAccess base class and custom versions of # hasValueForKey() and valueForKey(). This all means that UserDict's (such as # os.environ) are key/value accessible. # @@ 2000-05-07 ce: CGIWrapper.py duplicates this. def _enhanceUserDict(): from UserDict import UserDict if not NamedValueAccess in UserDict.__bases__: UserDict.__bases__ = UserDict.__bases__ + (NamedValueAccess,) def _UserDict_hasValueForKey(self, key): return self.has_key(key) def _UserDict_valueForKey(self, key, default=NoDefault): if default is NoDefault: if self.has_key(key): return self[key] else: raise ValueForKeyError, key else: return self.get(key, default) setattr(UserDict, 'hasValueForKey', _UserDict_hasValueForKey) setattr(UserDict, 'valueForKey', _UserDict_valueForKey) _enhanceUserDict()